Person-Oriented Approaches to Mediation Analysis

4^TH PAPER WITHIN ORGANIZED POSTER FORUM

ABSTRACT BODY: Two general approaches to statistical analyses have been proposed in the research literature. The dominant approach is focused on relations between variables and generally treats individual differences as ignorable random error. The second approach, which has received less attention, uses analyses focused on the differences between people, or latent groups of people, to ask how individuals differ across a constellation of relevant variables. The purpose of this presentation is to outline person-oriented methods and contrast them with variable-oriented methods in mediation analysis. Traditional variable oriented mediation assumes the population undergoes a homogenous reaction to the mediating process. However, mediation is also an intra-individual process where each person passes from a predictor, through a mediator, to an outcome. (Collins, Graham, & Flaherty, 1998). This view is supported by the existence of moderators of mediated intervention effects and the existence of individual differences in the relation of the mediator to outcome (MacKinnon & Pirlott, 2015). Focusing on the individual’s pattern of responses allows for individual differences in the mediating process, which may be detected using person-oriented statistical methods. A focus on individual differences in program effectiveness is a longstanding aspect of prevention science (Sloboda & Petras, 2014).

The description of person and variable oriented mediation methods is illustrated in the simplest case where each variable in the single-mediator model is binary, so there are eight possible patterns of responses: a person can either have or not have the predictor, either have or not have the mediator, and either have or not have the outcome. Variable and person-oriented approaches to mediation are described for these eight patterns of responses. We describe how each of these eight patterns does or does not correspond to a mediation process and describe person-oriented methods to mediation analysis including configural frequency mediation (von Eye, Mun, & Mair, 2009) and stage-sequential approaches (Collins et al., 1998). In this way the person versus variable-oriented approaches are contrasted.

A simulation study compares a person-oriented method, configural frequency mediation, with two variable-oriented methods: logistic regression and the causal mediation formula (Pearl, 2012). The study consists of two data generating models, the first corresponding to a variable oriented single mediator model, and the second corresponding to a heterogeneous population with groups of observations that have either full, partial, or no mediation. Results showing each method’s ability to detect full and partial mediation at varying levels of effect and sample size are discussed.